CGYRO Multicore and Accelerator Optimization

Project Description

As already discussed, the physics motivation for this effort has been to resolve outstanding issues in the dynamics of turbulence in the pedestal-top region. Specifically, we want to address persistent disagreements between GYRO (USA) and GENE (Germany), the two leading GK codes worldwide, on the DIII-D L-mode shortfall case. While the two codes have a long history of close agreement on a remarkably broad spectrum of operating regimes, in the shortfall regime the codes disagree by a factor of two. This disagreement has emerged as perhaps the most troubling aspect of core turbulence analysis in the past 5 years. Because full resolution of the turbulence spectrum up to electron scales may be critical for settling this controversy, we have used algorithms and array distribution strategies that we hope will perform optimally for multiscale simulations.Quite independently, for ITER operating conditions, recent work shows that realistic assessment of electron transport in fusion devices requires (massive) multi-scale simulations. It is well-known that a full multiscale simulation (presently, only GYRO and GENE are capable of such simulations) is a factor of 100-1000 more expensive than a traditional long-wavelength simulation, thus accounting for their rarity. Thus, to make these simulations routine, we believe a targeted effort (i.e., a new special-purpose code) is required. That is why CGYRO has beoen built from the ground up to address this regime.